Phys-Diff: A Physics-Inspired Latent Diffusion Model for Tropical Cyclone Forecasting
This work addresses the need for physically consistent predictions in tropical cyclone forecasting, which is critical for disaster warning and emergency response, representing a novel method for a known bottleneck.
The paper tackles the problem of tropical cyclone forecasting by proposing Phys-Diff, a physics-inspired latent diffusion model that disentangles latent features and uses cross-task attention to embed physical consistency, achieving state-of-the-art performance on global and regional datasets.
Tropical cyclone (TC) forecasting is critical for disaster warning and emergency response. Deep learning methods address computational challenges but often neglect physical relationships between TC attributes, resulting in predictions lacking physical consistency. To address this, we propose Phys-Diff, a physics-inspired latent diffusion model that disentangles latent features into task-specific components (trajectory, pressure, wind speed) and employs cross-task attention to introduce prior physics-inspired inductive biases, thereby embedding physically consistent dependencies among TC attributes. Phys-Diff integrates multimodal data including historical cyclone attributes, ERA5 reanalysis data, and FengWu forecast fields via a Transformer encoder-decoder architecture, further enhancing forecasting performance. Experiments demonstrate state-of-the-art performance on global and regional datasets.